[Feature Request] Add ElasticSearch Vector Store Support

[gkswns0531]

Do you need to file an issue?

• I have searched the existing issues and this feature is not already filed.
• My model is hosted on OpenAI or Azure. If not, please look at the "model providers" issue and don't file a new one here.
• I believe this is a legitimate feature request, not just a question. If this is a question, please use the Discussions area.

Is your feature request related to a problem? Please describe.

Summary

I would like to propose adding ElasticSearch vector store support to GraphRAG to enhance enterprise adoption and infrastructure compatibility. This feature would allow organizations with existing ElasticSearch deployments to leverage GraphRAG without requiring additional vector database infrastructure.

I’ve attached the PR for the ElasticSearch support I implemented. I would greatly appreciate it if you could review it. I tested the full indexing workflow using the CLI and confirmed that all four search methods—local, global, basic, and drift—are working correctly.

Related PR: #2002

Motivation

ElasticSearch is one of the most widely adopted vector search solutions in the enterprise RAG ecosystem. Many organizations have already invested significantly in ElasticSearch infrastructure for their search and analytics needs. Currently, GraphRAG supports LanceDB and other vector stores, but lacks ElasticSearch integration, which creates a barrier for enterprise adoption.

Key Benefits:

• Infrastructure Reuse: Organizations can leverage existing ElasticSearch clusters
• Enterprise Readiness: ElasticSearch offers enterprise-grade features like security, monitoring, and scaling
• Cost Efficiency: Reduces infrastructure complexity and operational overhead
• Industry Standard: Many RAG implementations in production already use ElasticSearch as their vector backend

Proposed Solution

I propose implementing an ElasticSearchVectorStore class that follows the existing BaseVectorStore interface pattern, similar to the current LanceDB implementation. This approach would ensure:

• Full compatibility with existing GraphRAG workflows
• Seamless integration with the current VectorStoreFactory pattern
• Consistent API surface across all vector store implementations
• No breaking changes to existing functionality

Technical Approach:

• Implement BaseVectorStore interface with ElasticSearch backend
• Support for KNN vector search using ElasticSearch's native capabilities
• Dynamic vector dimension detection (following LanceDB's flexibility)
• Bulk document operations for efficient indexing
• Proper error handling and connection management

Configuration Example:

vector_store:
  default_vector_store:
    type: elasticsearch
    url: "http://localhost:9200"
    index_name: "graphrag_vectors"
    # Additional ElasticSearch-specific configurations

Implementation Considerations

• Compatibility: The implementation would follow the exact same patterns as the existing LanceDB vector store
• Testing: Comprehensive integration tests to ensure feature parity
• Documentation: Clear setup and configuration guidelines
• Dependencies: Minimal additional dependencies (elasticsearch-py client)

[natoverse]

Main has fixed the typings for custom vector store registration, and includes a notebook demonstrating how to register your own store. This will be publisehd to pypi on the next release as well. We have to limit how many we officially support for maintenance reasons, but feel free to link to an implementation that others can use.